https://github.com/grafana/vectorapi

pgvector + embeddings API
https://github.com/grafana/vectorapi

document-retrieval embeddings llms pgvector

Last synced: 4 months ago
JSON representation

pgvector + embeddings API

• Host: GitHub
• URL: https://github.com/grafana/vectorapi
• Owner: grafana
• License: agpl-3.0
• Archived: true
• Created: 2023-08-30T14:59:06.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2023-12-14T13:00:07.000Z (about 2 years ago)
• Last Synced: 2025-01-29T11:52:08.036Z (about 1 year ago)
• Topics: document-retrieval, embeddings, llms, pgvector
• Language: Python
• Homepage: https://grafana.github.io/vectorapi/
• Size: 559 KB
• Stars: 19
• Watchers: 115
• Forks: 4
• Open Issues: 3
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# VectorAPI

[![License](https://img.shields.io/github/license/grafana/vectorapi)](LICENSE)

[![Build Status](https://drone.grafana.net/api/badges/grafana/vectorapi/status.svg)](https://drone.grafana.net/grafana/vectorapi)

[![Docker](https://img.shields.io/docker/v/grafana/vectorapi?logo=docker)](https://hub.docker.com/r/grafana/vectorapi/tags)

[![API Docs](https://img.shields.io/badge/docs-api-blue)](https://grafana.github.io/vectorapi/)

VectorAPI is a service for managing vector collections and performing vector similarity queries using a PostgreSQL vector database with the `pgvector` extension. Utilizes `fastapi` for the HTTP API, `pgvector` and SQLAlchemy for the vector database side and relies on `pytorch` for computing embeddings.

## Getting started

### Existing database

To get started with the VectorAPI, run:

```sh

docker run -p 8889:8889 -e DB_URL=postgresql+asyncpg://:@:/ grafana/vectorapi

```

### New database

You can bring up a postgres database (`ankane/pgvector`) and vectorapi instance using docker compose:

```sh

docker compose up --build

```

To populate the local DB instance with test data from HuggingFace (see [Grafana public datasets](https://huggingface.co/grafanalabs)) run:

```sh

make populate-db

```

## Making requests

See [API docs](https://grafana.github.io/vectorapi/) for more details.

### Embedding text

```sh

curl -X POST "http://localhost:8889/v1/embeddings" \

    -H "Content-Type: application/json" \

    -d '{"input":"I enjoy taking long walks along the beach.", "model":"BAAI/bge-small-en-v1.5"}'

```

### Adding a vector to a collection

1. Create a collection

```sh

curl -X POST "http://localhost:8889/v1/collections/create" \

    -H "Content-Type: application/json" \

    -d '{"collection_name":"my_collection", "dimension":384}'

```

2. Add a vector to the collection

```sh

curl -X POST "http://localhost:8889/v1/collections/my_collection/upsert" \

    -H "Content-Type: application/json" \

    -d '{"id":"abc1", "metadata":{"key":"value"}, "input":"I enjoy taking long walks along the beach."}'

```

### Vector search

```sh

curl -X POST "http://localhost:8889/v1/collections/my_collection/search" \

    -H "Content-Type: application/json" \

    -d '{"input":"beach walks"}'

```